Search

This paper demonstrates RAGE, an interactive tool for explaining Large Language Models (LLMs) augmented with retrieval capabilities; i.e., able to query external sources and pull relevant information into their input context. Our explanations are counterfactual in the sense that they identify parts of the input context that, when removed, change the answer to the question posed to the LLM. RAGE includes pruning methods to navigate the vast space of possible explanations, allowing users to view the provenance of the produced answers., Comment: Accepted by ICDE 2024 (Demonstration Track)

We study committee elections from a perspective of finding the most conflicting candidates, that is, candidates that imply the largest amount of conflict, as per voter preferences. By proposing basic axioms to capture this objective, we show that none of the prominent multiwinner voting rules meet them. Consequently, we design committee voting rules compliant with our desiderata, introducing conflictual voting rules. A subsequent deepened analysis sheds more light on how they operate. Our investigation identifies various aspects of conflict, for which we come up with relevant axioms and quantitative measures, which may be of independent interest. We support our theoretical study with experiments on both real-life and synthetic data., Comment: Accepted for publication at IJCAI-24; 27 pages; 11 figures

Quantization-aware training comes with a fundamental challenge: the derivative of quantization functions such as rounding are zero almost everywhere and nonexistent elsewhere. Various differentiable approximations of quantization functions have been proposed to address this issue. In this paper, we prove that when the learning rate is sufficiently small, a large class of weight gradient estimators is equivalent with the straight through estimator (STE). Specifically, after swapping in the STE and adjusting both the weight initialization and the learning rate in SGD, the model will train in almost exactly the same way as it did with the original gradient estimator. Moreover, we show that for adaptive learning rate algorithms like Adam, the same result can be seen without any modifications to the weight initialization and learning rate. We experimentally show that these results hold for both a small convolutional model trained on the MNIST dataset and for a ResNet50 model trained on ImageNet.

Combining model-based and model-free reinforcement learning approaches, this paper proposes and analyzes an $\epsilon$-policy gradient algorithm for the online pricing learning task. The algorithm extends $\epsilon$-greedy algorithm by replacing greedy exploitation with gradient descent step and facilitates learning via model inference. We optimize the regret of the proposed algorithm by quantifying the exploration cost in terms of the exploration probability $\epsilon$ and the exploitation cost in terms of the gradient descent optimization and gradient estimation errors. The algorithm achieves an expected regret of order $\mathcal{O}(\sqrt{T})$ (up to a logarithmic factor) over $T$ trials.

Variational quantum computing offers a flexible computational paradigm with applications in diverse areas. However, a key obstacle to realizing their potential is the Barren Plateau (BP) phenomenon. When a model exhibits a BP, its parameter optimization landscape becomes exponentially flat and featureless as the problem size increases. Importantly, all the moving pieces of an algorithm -- choices of ansatz, initial state, observable, loss function and hardware noise -- can lead to BPs when ill-suited. Due to the significant impact of BPs on trainability, researchers have dedicated considerable effort to develop theoretical and heuristic methods to understand and mitigate their effects. As a result, the study of BPs has become a thriving area of research, influencing and cross-fertilizing other fields such as quantum optimal control, tensor networks, and learning theory. This article provides a comprehensive review of the current understanding of the BP phenomenon., Comment: 21 pages, 10 boxes

We report on our experiences replicating 7.3 petabytes (PB) of Earth System Grid Federation (ESGF) climate simulation data from Lawrence Livermore National Laboratory (LLNL) in California to Argonne National Laboratory (ANL) in Illinois and Oak Ridge National Laboratory (ORNL) in Tennessee. This movement of some 29 million files, twice, undertaken in order to establish new ESGF nodes at ANL and ORNL, was performed largely automatically by a simple replication tool, a script that invoked Globus to transfer large bundles of files while tracking progress in a database. Under the covers, Globus organized transfers to make efficient use of the high-speed Energy Sciences network (ESnet) and the data transfer nodes deployed at participating sites, and also addressed security, integrity checking, and recovery from a variety of transient failures. This success demonstrates the considerable benefits that can accrue from the adoption of performant data replication infrastructure.

(abridged) Protostellar outflows exhibit large variations in their structure depending on the observed gas emission. This study analyzes the atomic jet and molecular outflow in the Class I protostar, TMC1A to characterize morphology and identify previously undetected spatial features with JWST's NIRSpec IFU. In addition to identifying a large number of Fe II and H2 lines, we have detected the bipolar Fe jet by revealing, for the first time, the presence of a red-shifted atomic jet. Similarly, the red-shifted component of the H2 slower wide-angle outflow is observed. Both Fe II and H2 red-shifted emission exhibit significantly lower flux densities compared to their blue-shifted counterparts. Additionally, we report the detection of a collimated high-velocity (100 km s-1), blue-shifted H2 outflow, suggesting the presence of a molecular jet in addition to the well-known wider angle low-velocity structure. The Fe II and H2 jets show multiple intensity peaks along the jet axis, which may be associated with ongoing or recent outburst events. In addition to the variation in their intensities, the H2 wide-angle outflow exhibits a "ring"-like structure. The blue-shifted H2 outflow also shows a left-right brightness asymmetry likely due to interactions with the surrounding ambient medium and molecular outflows. Using the Fe II line ratios, the extinction along the atomic jet is estimated to be between Av = 10-30 on the blue-shifted side, with a trend of decreasing extinction with distance from the protostar. A similar Av is found for the red-shifted side, supporting the argument for an intrinsic red-blue outflow lobe asymmetry rather than environmental effects such as extinction. This intrinsic difference revealed by the unprecedented sensitivity of JWST, suggests that younger outflows already exhibit the red-blue side asymmetry more commonly observed towards jets associated with Class II disks., Comment: 16 pages, 11 figures

In the online disjoint set covers problem, the edges of a hypergraph are revealed online, and the goal is to partition them into a maximum number of disjoint set covers. That is, n nodes of a hypergraph are given at the beginning, and then a sequence of hyperedges (subsets of [n]) is presented to an algorithm. For each hyperedge, an online algorithm must assign a color (an integer). Once an input terminates, the gain of the algorithm is the number of colors that correspond to valid set covers (i.e., the union of hyperedges that have that color contains all n nodes). We present a deterministic online algorithm that is O(log^2 n)-competitive, exponentially improving on the previous bound of O(n) and matching the performance of the best randomized algorithm by Emek et al. [ESA 2019]. For color selection, our algorithm uses a novel potential function, which can be seen as an online counterpart of the derandomization method of conditional probabilities and pessimistic estimators. There are only a few cases where derandomization has been successfully used in the field of online algorithms. In contrast to previous approaches, our result extends this tool to tackle the following new challenges: (i) the potential function derandomizes not only the Chernoff bound, but also the coupon collector's problem, (ii) the value of OPT of the maximization problem is not bounded a priori, and (iii) we do not produce a fractional solution first, but work directly on the input.

This work aims to utilize the increased sensitivity and resolution of the JWST to quantify the HDO/H$_{2}$O ratio in ices toward young stellar objects (YSOs) and to determine if the HDO/H$_{2}$O ratios measured in the gas phase toward massive YSOs (MYSOs) are representative of the ratios in their ice envelopes. Two protostars observed in the Investigating Protostellar Accretion (IPA) program using JWST NIRSpec were analyzed: HOPS 370, an intermediate-mass YSO (IMYSO), and IRAS 20126+4104, a MYSO. The HDO ice toward these sources was detected above the 3$\sigma$ level and quantified via its 4.1 $\mu$m band. The contributions from the CH$_{3}$OH combination modes to the observed optical depth in this spectral region were constrained via the CH$_{3}$OH 3.53 $\mu$m band to ensure that the integrated optical depth of the HDO feature was not overestimated. H$_{2}$O ice was quantified via its 3 $\mu$m band. From these fits, ice HDO/H$_{2}$O abundance ratios of 4.6$\pm$1.8$\times$10$^{-3}$ and 2.6$\pm$1.2$\times$10$^{-3}$ are obtained for HOPS 370 and IRAS 20126+4104, respectively. The simultaneous detections of both crystalline HDO and crystalline H$_{2}$O corroborate the assignment of the observed feature at 4.1 $\mu$m to HDO ice. The ice HDO/H$_{2}$O ratios are similar to the highest reported gas HDO/H$_{2}$O ratios measured toward MYSOs as well as the hot inner regions of isolated low-mass protostars, suggesting that at least some of the gas HDO/H$_{2}$O ratios measured toward massive hot cores are representative of the HDO/H$_{2}$O ratios in ices. The need for an H$_{2}$O-rich CH$_{3}$OH component in the CH$_{3}$OH ice analysis supports recent experimental and observational results that indicate that some CH$_{3}$OH ice may form prior to the CO freeze-out stage in H$_{2}$O-rich ice layers., Comment: Accepted for publication in A&A. 23 pages, 17 figures, 10 tables

With 5394 security certificates of IT products and systems, the Common Criteria for Information Technology Security Evaluation have bred an ecosystem entangled with various kind of relations between the certified products. Yet, the prevalence and nature of dependencies among Common Criteria certified products remains largely unexplored. This study devises a novel method for building the graph of references among the Common Criteria certified products, determining the different contexts of references with a supervised machine-learning algorithm, and measuring how often the references constitute actual dependencies between the certified products. With the help of the resulting reference graph, this work identifies just a dozen of certified components that are relied on by at least 10% of the whole ecosystem -- making them a prime target for malicious actors. The impact of their compromise is assessed and potentially problematic references to archived products are discussed.

Internet of Things systems exists in various areas of our everyday life. For example, sensors installed in smart cities and homes are processed in edge and cloud computing centres providing several benefits that improve our lives. The place of data processing is related to the required system response times -- processing data closer to its source results in a shorter system response time. The Osmotic Computing concept enables flexible deployment of data processing services and their possible movement, just like particles in the osmosis phenomenon move between regions of different densities. At the same time, the impact of complex computer architecture on the environment is increasingly being compensated by the use of renewable and low-carbon energy sources. However, the uncertainty of supplying green energy makes the management of Osmotic Computing demanding, and therefore their autonomy is desirable. In the paper, we present a framework enabling osmotic computing simulation based on renewable energy sources and autonomic osmotic agents, allowing the analysis of distributed management algorithms. We discuss the challenges posed to the framework and analyze various management algorithms for cooperating osmotic agents. In the evaluation we show that changing the adaptation logic of the osmotic agents, it is possible to increase the self-consumption of renewable energy sources or increase the usage of low emission ones.

This thesis pertains to the study of elliptic and parabolic partial differential equations on "thin" structures. The first main objective is to establish the strong and weak low-dimensional counterparts of the parabolic Neumann problem. The main technical result is proving the closedness of the low-dimensional second-order operator. To construct a semigroup, a variant of Magyar of the Hille-Yosida Theorem for non-invertible operators is adapted. An alternative direction of study is presented to extend the class of accessible initial data. Weak-type parabolic problems are defined, and the existence of solutions is obtained by the application of the Lions version of the Lax-Milgram Lemma. The second aspect of the thesis is to examine the higher regularity of weak solutions to low-dimensional elliptic problems. We prove that for any weak low-dimensional elliptic solution, some Cosserat vector field exists as a witness of the membership of the solution to the domain of the second-order operator., Comment: PhD thesis

Using the MIRI/MRS spectrometer on JWST, we have detected pure rotational, suprathermal OH emissions from the vicinity of the intermediate-mass protostar HOPS 370 (OMC2/FIR3). These emissions are observed from shocked knots in a jet/outflow, and originate in states of rotational quantum number as high as 46 that possess excitation energies as large as $E_U/k = 4.65 \times 10^4$ K. The relative strengths of the observed OH lines provide a powerful diagnostic of the ultraviolet radiation field in a heavily-extinguished region ($A_V \sim 10 - 20$) where direct UV observations are impossible. To high precision, the OH line strengths are consistent with a picture in which the suprathermal OH states are populated following the photodissociation of water in its $\tilde B - X$ band by ultraviolet radiation produced by fast ($\sim 80\,\rm km\,s^{-1}$) shocks along the jet. The observed dominance of emission from symmetric ($A^\prime$) OH states over that from antisymmetric ($A^{\prime\prime}$) states provides a distinctive signature of this particular population mechanism. Moreover, the variation of intensity with rotational quantum number suggests specifically that Ly$\alpha$ radiation is responsible for the photodissociation of water, an alternative model with photodissociation by a 10$^4$ K blackbody being disfavored at a high level of significance. Using measurements of the Br$\alpha$ flux to estimate the Ly$\alpha$ production rate, we find that $\sim 4\%$ of the Ly$\alpha$ photons are absorbed by water. Combined with direct measurements of water emissions in the $\nu_2 = 1 -0$ band, the OH observations promise to provide key constraints on future models for the diffusion of Ly$\alpha$ photons in the vicinity of a shock front., Comment: 30 pages, 7 figures. Accepted for publication in ApJ Letters

We present a coarse-grained C$\alpha$-based protein model that can be used to simulate structured, intrinsically disordered and partially disordered proteins. We use a Go-like potential for the structured parts and two different variants of a transferable potential for the disordered parts. The first variant uses dynamic structure-based (DSB) contacts that form and disappear quasi-adiabatically during the simulation. By using specific structural criteria we distinguish sidechain-sidechain, sidechain-backbone and backbone-backbone contacts. The second variant is a non-radial multi-body pseudo-improper-dihedral (PID) potential that does not include time-dependent terms but requires more computational resources. Our model can simulate in reasonable time thousands of residues on millisecond time scales.

Time series methods are of fundamental importance in virtually any field of science that deals with temporally structured data. Recently, there has been a surge of deterministic transformer models with time series-specific architectural biases. In this paper, we go in a different direction by introducing tsGT, a stochastic time series model built on a general-purpose transformer architecture. We focus on using a well-known and theoretically justified rolling window backtesting and evaluation protocol. We show that tsGT outperforms the state-of-the-art models on MAD and RMSE, and surpasses its stochastic peers on QL and CRPS, on four commonly used datasets. We complement these results with a detailed analysis of tsGT's ability to model the data distribution and predict marginal quantile values.

A powerful and robust control system is a crucial, often neglected, pillar of any modern, complex physics experiment that requires the management of a multitude of different devices and their precise time synchronisation. The AEgIS collaboration presents CIRCUS, a novel, autonomous control system optimised for time-critical experiments such as those at CERN's Antiproton Decelerator and, more broadly, in atomic and quantum physics research. Its setup is based on Sinara/ARTIQ and TALOS, integrating the ALPACA analysis pipeline, the last two developed entirely in AEgIS. It is suitable for strict synchronicity requirements and repeatable, automated operation of experiments, culminating in autonomous parameter optimisation via feedback from real-time data analysis. CIRCUS has been successfully deployed and tested in AEgIS; being experiment-agnostic and released open-source, other experiments can leverage its capabilities.

Large language models (LLMs) have recently been used as backbones for recommender systems. However, their performance often lags behind conventional methods in standard tasks like retrieval. We attribute this to a mismatch between LLMs' knowledge and the knowledge crucial for effective recommendations. While LLMs excel at natural language reasoning, they cannot model complex user-item interactions inherent in recommendation tasks. We propose bridging the knowledge gap and equipping LLMs with recommendation-specific knowledge to address this. Operations such as Masked Item Modeling (MIM) and Bayesian Personalized Ranking (BPR) have found success in conventional recommender systems. Inspired by this, we simulate these operations through natural language to generate auxiliary-task data samples that encode item correlations and user preferences. Fine-tuning LLMs on such auxiliary-task data samples and incorporating more informative recommendation-task data samples facilitates the injection of recommendation-specific knowledge into LLMs. Extensive experiments across retrieval, ranking, and rating prediction tasks on LLMs such as FLAN-T5-Base and FLAN-T5-XL show the effectiveness of our technique in domains such as Amazon Toys & Games, Beauty, and Sports & Outdoors. Notably, our method outperforms conventional and LLM-based baselines, including the current SOTA, by significant margins in retrieval, showcasing its potential for enhancing recommendation quality., Comment: Accepted to the NAACL 2024 Findings

Low-precision quantization is recognized for its efficacy in neural network optimization. Our analysis reveals that non-quantized elementwise operations which are prevalent in layers such as parameterized activation functions, batch normalization, and quantization scaling dominate the inference cost of low-precision models. These non-quantized elementwise operations are commonly overlooked in SOTA efficiency metrics such as Arithmetic Computation Effort (ACE). In this paper, we propose ACEv2 - an extended version of ACE which offers a better alignment with the inference cost of quantized models and their energy consumption on ML hardware. Moreover, we introduce PikeLPN, a model that addresses these efficiency issues by applying quantization to both elementwise operations and multiply-accumulate operations. In particular, we present a novel quantization technique for batch normalization layers named QuantNorm which allows for quantizing the batch normalization parameters without compromising the model performance. Additionally, we propose applying Double Quantization where the quantization scaling parameters are quantized. Furthermore, we recognize and resolve the issue of distribution mismatch in Separable Convolution layers by introducing Distribution-Heterogeneous Quantization which enables quantizing them to low-precision. PikeLPN achieves Pareto-optimality in efficiency-accuracy trade-off with up to 3X efficiency improvement compared to SOTA low-precision models., Comment: Accepted in CVPR 2024. 10 Figures, 9 Tables

The origin and impact of outflows on proto-planetary disks and planet formation are key open questions. DG Tau B, a Class I protostar with a structured disk and a striking rotating conical CO outflow, recently identified with ALMA as one of the best MHD disk wind candidate, is an ideal target for studying these phenomena. Our aim is to analyse the outflow components intermediate between the fast axial jet and the wider molecular CO outflow to discriminate between the different scenarios at their origin (irradiated/shocked disk wind or swept-up material). Using observations from JWST NIRSpec-IFU, NIRCam and SINFONI/VLT, we investigate the morphology, kinematics and excitation conditions of H$_2$ emission lines of the red-shifted outflow lobe. We find an onion-like structure of the outflows with increasing temperature, velocity and collimation towards the flow axis. The red-shifted H$_2$ emission reveals a narrow conical cavity nested inside the CO outflow and originating from the inner disk regions (< 6 au). The H$_2$ shell exhibits a constant vertical velocity ($\simeq$22 km/s), twice faster that of the CO flow and an average mass flux of $\dot{M}$(H$_2$) = 3e-11 M$_\odot$/yr significantly lower than the jet and CO values, suggesting low H$_2$ abundance. The global layered structure of the H$_2$/CO outflows is consistent with an MHD disk wind scenario, with the hot H$_2$ possibly tracing an inner dense photodissociation layer of the wind coming from a launching radius in the disk of 0.2-0.4 au. Further analysis, including MIRI observations will provide additional insights into the H$_2$ excitation mechanisms and the origin of the layered outflows observed in DG Tau B., Comment: 23 pages, 21 figures, 2 tables, submitted and reviewed to A&A

The ALICE experiment at the LHC measures properties of the strongly interacting matter formed in ultrarelativistic heavy-ion collisions. Such studies require accurate particle identification (PID). ALICE provides PID information via several detectors for particles with momentum from about 100 MeV/c up to 20 GeV/c. Traditionally, particles are selected with rectangular cuts. A much better performance can be achieved with machine learning (ML) methods. Our solution uses multiple neural networks (NN) serving as binary classifiers. Moreover, we extended our particle classifier with Feature Set Embedding and attention in order to train on data with incomplete samples. We also present the integration of the ML project with the ALICE analysis software, and we discuss domain adaptation, the ML technique needed to transfer the knowledge between simulated and real experimental data., Comment: Proceedings of 3rd Artificial Intelligence for the Electron-Ion Collider workshop -- AI4EIC2023, 28.11-1.12.2023

Geomagnetic storms occurring due to sustained, high-speed solar winds are known to induce currents in power distribution networks. These geomagnetically induced currents (GICs) can cause high voltage transformers (HVT) to overheat, thus resulting in a catastrophic electricity loss event (CELE). Since significant portions of infrastructures around the world rely heavily on access to electric power, it is essential to estimate the risks associated with GICs on a global scale. We assemble multiple methodologies across various scientific disciplines to develop a framework assessing the probability of a severe geomagnetic storm causing a long-term, widespread power outage. Our model incorporates thermal models of HVT tie bar hot spots, historical geoelectric field estimates, and a global conductivity model to estimate the risk of long-term power outage for regions between -70 degrees and 80 degrees geomagnetic latitude due to transformer overheating failure. Assuming a uniform 33% HVT spare capacity, our analysis indicates that a 1 in 10,000 year storm would result in approximately 1% of the population in Europe and North America experiencing a long-term (months to years) electricity loss.

In this paper we study predictive mean matching mass imputation estimators to integrate data from probability and non-probability samples. We consider two approaches: matching predicted to predicted ($\hat{y}-\hat{y}$~matching; PMM A) and predicted to observed ($\hat{y}-y$~matching; PMM B) values. We prove the consistency of two semi-parametric mass imputation estimators based on these approaches and derive their variance and estimators of variance. We underline the differences of our approach with the nearest neighbour approach proposed by Yang et al. (2021) and prove consistency of the PMM A estimator under model mis-specification. Our approach can be employed with non-parametric regression techniques, such as kernel regression, and the analytical expression for variance can also be applied in nearest neighbour matching for non-probability samples. We conduct extensive simulation studies in order to compare the properties of this estimator with existing approaches, discuss the selection of $k$-nearest neighbours, and study the effects of model mis-specification. The paper finishes with empirical study in integration of job vacancy survey and vacancies submitted to public employment offices (admin and online data). Open source software is available for the proposed approaches., Comment: 52 pages; codes: https://github.com/ncn-foreigners/paper-nonprob-pmm

Context. Gaia18cjb is one of the Gaia-alerted eruptive young star candidates which has been experiencing a slow and strong brightening during the last 13 years, similar to some FU Orionis-type objects. Aims. The aim of this work is to derive the young stellar nature of Gaia18cjb, determine its physical and accretion properties to classify its variability. Methods. We conducted monitoring observations using multi-filter optical and near-infrared photometry, as well as near-infrared spectroscopy. We present the analysis of pre-outburst and outburst optical and infrared light curves, color-magnitude diagrams in different bands, the detection of near-IR spectral lines, and estimates of both stellar and accretion parameters during the burst. Results. The optical light curve shows an unusually long (8 years) brightening event of 5 mag in the last 13 years, before reaching a plateau indicating that the burst is still on-going, suggesting a FUor-like nature. The same outburst is less strong in the infrared light curves. The near-infrared spectra, obtained during the outburst, exhibit emission lines typical of highly accreting low-intermediate mass young stars with typical EXor features. The spectral index of Gaia18cjb SED classifies it as a Class I in the pre-burst stage and a Flat Spectrum young stellar object (YSO) during the burst. Conclusions. Gaia18cjb is an eruptive YSO which shows FUor-like photometric features (in terms of brightening amplitude and length of the burst) and EXor-like spectroscopic features and accretion rate, as V350 Cep and V1647 Ori, classified as objects in between FUors and EXors, Comment: 21 pages, 13 figures

Remote sensing (RS) applications in the space domain demand machine learning (ML) models that are reliable, robust, and quality-assured, making red teaming a vital approach for identifying and exposing potential flaws and biases. Since both fields advance independently, there is a notable gap in integrating red teaming strategies into RS. This paper introduces a methodology for examining ML models operating on hyperspectral images within the HYPERVIEW challenge, focusing on soil parameters' estimation. We use post-hoc explanation methods from the Explainable AI (XAI) domain to critically assess the best performing model that won the HYPERVIEW challenge and served as an inspiration for the model deployed on board the INTUITION-1 hyperspectral mission. Our approach effectively red teams the model by pinpointing and validating key shortcomings, constructing a model that achieves comparable performance using just 1% of the input features and a mere up to 5% performance loss. Additionally, we propose a novel way of visualizing explanations that integrate domain-specific information about hyperspectral bands (wavelengths) and data transformations to better suit interpreting models for hyperspectral image analysis., Comment: 14 pages, 9 figures, ICLR 2024 Machine Learning for Remote Sensing (ML4RS) Workshop

Partial Multi-label Learning (PML) is a type of weakly supervised learning where each training instance corresponds to a set of candidate labels, among which only some are true. In this paper, we introduce \our{}, a novel probabilistic approach to this problem that extends the binary cross entropy to the PML setup. In contrast to existing methods, it does not require suboptimal disambiguation and, as such, can be applied to any deep architecture. Furthermore, experiments conducted on artificial and real-world datasets indicate that \our{} outperforms existing approaches, especially for high noise in a candidate set., Comment: Accepted to the International Conference on Data Science and Advanced Analytics (DSAA 2023)

With the advancements of transformer-based architectures, we observe the rise of natural language preprocessing (NLPre) tools capable of solving preliminary NLP tasks (e.g. tokenisation, part-of-speech tagging, dependency parsing, or morphological analysis) without any external linguistic guidance. It is arduous to compare novel solutions to well-entrenched preprocessing toolkits, relying on rule-based morphological analysers or dictionaries. Aware of the shortcomings of existing NLPre evaluation approaches, we investigate a novel method of reliable and fair evaluation and performance reporting. Inspired by the GLUE benchmark, the proposed language-centric benchmarking system enables comprehensive ongoing evaluation of multiple NLPre tools, while credibly tracking their performance. The prototype application is configured for Polish and integrated with the thoroughly assembled NLPre-PL benchmark. Based on this benchmark, we conduct an extensive evaluation of a variety of Polish NLPre systems. To facilitate the construction of benchmarking environments for other languages, e.g. NLPre-GA for Irish or NLPre-ZH for Chinese, we ensure full customization of the publicly released source code of the benchmarking system. The links to all the resources (deployed platforms, source code, trained models, datasets etc.) can be found on the project website: https://sites.google.com/view/nlpre-benchmark., Comment: Accepted at LREC-COLING 2024

Dynamic spectrum access is essential for radiocommunication and its limited spectrum resources. The key element of dynamic spectrum access systems is effective spectrum occupancy detection. In many cases, machine learning algorithms improve detection effectiveness. Because of the recent trend of using federated learning, a federated learning algorithm is presented in the context of distributed spectrum occupancy detection. The results of the work presented in the paper are based on actual signal samples collected in the laboratory. The proposed algorithm is effective, especially in the context of a set of sensors with faulty sensors., Comment: 4 pages, 2 figures

We introduce GUIDE, a novel continual learning approach that directs diffusion models to rehearse samples at risk of being forgotten. Existing generative strategies combat catastrophic forgetting by randomly sampling rehearsal examples from a generative model. Such an approach contradicts buffer-based approaches where sampling strategy plays an important role. We propose to bridge this gap by incorporating classifier guidance into the diffusion process to produce rehearsal examples specifically targeting information forgotten by a continuously trained model. This approach enables the generation of samples from preceding task distributions, which are more likely to be misclassified in the context of recently encountered classes. Our experimental results show that GUIDE significantly reduces catastrophic forgetting, outperforming conventional random sampling approaches and surpassing recent state-of-the-art methods in continual learning with generative replay.

This contribution presents the recent research developments within the Cosmic Ray Extremely Distributed Observatory (CREDO) in the search for resolution of various scientific puzzles, ranging from fundamental physical questions to applications like the determination of earthquake precursors. The state-of-the art theoretical, numerical and computational aspects of these phenomena are addressed, as well as recent experimental developments for detection., Comment: 9 pages, 7 figures. Based on the contribution to the XVIII Mexican Workshop on Particles and Fields in Puebla, Mexico, November 2022. M\'exico

In this work we are focused on the existence of Morse functions on a closed manifold $M$ which are far from being ordered, i.e. whose Reeb graphs have positive first Betti number, especially the maximal possible, equals $\operatorname{corank}(\pi_1(M))$. In the case of $3$-manifolds we describe the minimal number of critical points needed to construct such functions, which is related with the number of vertices of degree $2$ in Reeb graphs. We define a new invariant of $3$-manifold groups and their presentations, and using Heegaard splittings we show its utility in determining occurrence of disordered Morse functions. In particular, the Freiheitssatz, a result for one-relator groups, allows us to calculate this invariant in the case of orientable circle-bundles over a surface, which provides an interesting example of the behaviour of Morse functions., Comment: 13 pages

Various quantum algorithms require usage of arbitrary diagonal operators as subroutines. For their execution on a physical hardware, those operators must be first decomposed into target device's native gateset and its qubit connectivity for entangling gates. Here, we assume that the allowed gates are exactly the CX gate and the parameterized phase gate. We introduce a framework for the analysis of CX-only circuits and through its lens provide solution constructions for several different device topologies (fully-connected, linear and circular). We also introduce two additional variants of the problem. Those variants can be used in place of exact decomposition of the diagonal operator when the circuit following it satisfies a set of prerequisites, enabling further reduction in the CX cost of implementation. Finally, we discuss how to exploit the framework for the decomposition of a particular, rather than general, diagonal operator., Comment: 17 pages, 12 figures, 1 table

A basic primitive in quantum information is the computation of the moments $\mathbb{E}_U[{\rm Tr}[U\rho U^\dagger O]^t]$. These describe the distribution of expectation values obtained by sending a state $\rho$ through a random unitary $U$, sampled from some distribution, and measuring the observable $O$. While the exact calculation of these moments is generally hard, if $U$ is composed of local random gates, one can estimate $\mathbb{E}_U[{\rm Tr}[U\rho U^\dagger O]^t]$ by performing Monte Carlo simulations of a Markov chain-like process. However, this approach can require a prohibitively large number of samples, or suffer from the sign problem. In this work, we instead propose to estimate the moments via tensor networks, where the local gates moment operators are mapped to small dimensional tensors acting on their local commutant bases. By leveraging representation theoretical tools, we study the local tensor dimension and we provide bounds for the bond dimension of the matrix product states arising from deep circuits. We compare our techniques against Monte Carlo simulations, showing that we can significantly out-perform them. Then, we showcase how tensor networks can exactly compute the second moment when $U$ is a quantum neural network acting on thousands of qubits and having thousands of gates. To finish, we numerically study the anticoncentration phenomena of circuits with orthogonal random gates, a task which cannot be studied via Monte Carlo due to sign problems., Comment: 15 + 8 pages, 9 figures, updated to published version

The fractional material derivative appears as the fractional operator that governs the dynamics of the scaling limits of L\'evy walks - a stochastic process that originates from the famous continuous-time random walks. It is usually defined as the Fourier-Laplace multiplier, therefore, it can be thought of as a pseudo-differential operator. In this paper, we show that there exists a local representation in time and space, pointwise, of the fractional material derivative. This allows us to define it on a space of locally integrable functions which is larger than the original one in which Fourier and Laplace transform exist as functions. We consider several typical differential equations involving the fractional material derivative and provide conditions for their solutions to exist. In some cases, the analytical solution can be found. For the general initial value problem, we devise a finite volume method and prove its stability, convergence, and conservation of probability. Numerical illustrations verify our analytical findings. Moreover, our numerical experiments show superiority in the computation time of the proposed numerical scheme over a Monte Carlo method applied to the problem of probability density function's derivation.

In terms of complex radio environments especially in dense urban areas, a very interesting topic is considered - the utilization of reconfigurable intelligent surfaces. Basically, based on simple controls of the angle of reflection of the signal from the surface, it is possible to achieve different effects in a radio communication system. Maximizing or minimizing the received power at specific locations near the reflecting surface is the most important effect. Thanks to this, it is possible to: receive a signal in a place where it was not possible, detect spectrum occupancy in a place where the sensor could not make a correct detection, or minimize interference in a specific receiver. In this paper, all three concepts are presented, and, using a simple ray tracing simulation, the potential profit in each scenario is shown. In addition, a scenario was analyzed in which several of the aforementioned situations are combined., Comment: 4 pages, 7 figures, English language

Reconfigurable intelligent surfaces can be successfully used to control the radio environment. Simple control of the reflection angle of the signal from the surface allows maximization or minimization of the received power in specific places. The paper presents simulations where it is possible to receive a signal in a place where it was not possible, to detect the occupancy of the spectrum in a place where the sensor was unable to make correct detection or to minimize interference in a specific receiver., Comment: 4 pages, in Polish language, 5 figures, presented during conference

Approaches to aligning large language models (LLMs) with human values has focused on correcting misalignment that emerges from pretraining. However, this focus overlooks another source of misalignment: bad actors might purposely fine-tune LLMs to achieve harmful goals. In this paper, we present an emerging threat model that has arisen from alignment circumvention and fine-tuning attacks. However, lacking in previous works is a clear presentation of the conditions for effective defence. We propose a set of conditions for effective defence against harmful fine-tuning in LLMs called "Immunization conditions," which help us understand how we would construct and measure future defences. Using this formal framework for defence, we offer a synthesis of different research directions that might be persued to prevent harmful fine-tuning attacks and provide a demonstration of how to use these conditions experimentally showing early results of using an adversarial loss to immunize LLama2-7b-chat.